The first figure displays the performance as measured by size of each HC * Test combination at the .05 level for each sample size. Each box plot represents error distribution (3) * error structure (7) * variable (6). HC1 combined with the model version of saddlepoint seems to outperform all other combinations on average.